Welcome to the newest edition of ID Tech’s AI Update. Here’s the latest big news on the shifting landscape of AI and identity technology:
A major US-UK AI partnership may be in the works, with reports that OpenAI and Nvidia are preparing to commit billions of dollars to AI infrastructure projects in Britain in partnership with London-based Nscale Global Holdings. The plan is expected to be unveiled during US President Donald Trump’s planned visit to the country next week.
Microsoft has signed a five-year GPU compute agreement with AI infrastructure firm Nebius, with the arrangement initially priced at $17.4 billion, and an option to scale up to $20 billion depending on demand. Nebius was spun out of Yandex, a major Russian internet company, and now offers things like GPU clusters, storage, and cloud software on an international scale. The deal will involve establishing capacity from a new facility in Vineland, New Jersey.
Microsoft is also making a deal to integrate AI from OpenAI rival Anthropic into its Office 365 and Microsoft 365 suites. While Microsoft is by far OpenAI’s biggest investor, it seems to be taking measures to ensure that not all of its eggs are in one basket. Internal Microsoft testing reportedly found that Claude Sonnet 4 outperforms OpenAI models in certain specific tasks, such as making PowerPoint slides and automating financial functions in Excel.
Tencent Holdings has hired a former OpenAI researcher as it seeks to scale up its AI expertise. Having earned a PhD in computer science at Princeton, Yao Shunyu has been focused largely on AI agents, with about 10 published papers to his name. Tencent is reported to have offered him 100 million yuan, or roughly $13-15 million, illustrating how the frothy competition for AI talent is not a phenomenon confined to Silicon Valley.
Reflection AI is in talks to raise about $1 billion at a valuation of roughly $4.5 to $5.5 billion – a nearly tenfold increase from six months ago, when it was valued at $545 million. The startup was founded by ex-DeepMind/Google AI researchers Misha Laskin and Ioannis Antonoglou just over a year ago, and is working on AI-driven coding tools in the short term, with an eye to “superintelligence” in the long term.
Shanghai Ant Lingbo Technology Co. (or “Robbyant”) has unveiled its first humanoid robot, dubbed R1, and says it is already in mass production. For those watching the frontier of the robotics space, it isn’t all that impressive: The robot has arms but wheels instead of bipedal legs, and moves somewhat slowly. But it’s notable for its pedigree. Robbyant is the robotics arm of Ant Group, a gigantic Chinese fintech firm affiliated with the billionaire Jack Ma. Why is a fintech firm getting into robotics? Who knows; but this is the strange future that AI is ushering in.
Korea’s SK Hynix has achieved a memory breakthrough in AI chips. The firm’s new HBM4 (High-Bandwidth Memory generation 4) roughly doubles the bandwidth of previous HBM stacks, has data transfer speeds of up to about 10 GT/s (gigatransfers per second) per pin, and improves power efficiency by over 40 percent. Memory bandwidth and efficiency are critical bottlenecks in large AI models and AI accelerators, with communications between memory and compute being a more challenging limit than raw compute power, so this breakthrough could help to make AI processing considerably faster and cheaper.
Nvidia has unveiled a new class of GPUs purpose-built for generative video and complex software tasks. Its Rubin CPX is tailored for massive-context inference and can process data from huge context windows – that is, the amount of information inputted into the system in order to generate a result. Nvidia says the solution delivers a large multiple of performance over its prior generation, effectively offering more powerful compute to video generation platforms, among other potential clients. It will be available in late 2026 as part of a rack-scale platform called Vera Rubin NVL144 CPX.
ASML has invested €1.3 billion in French AI startup Mistral as part of a larger €1.7 billion Series C funding round, giving ASML a roughly 11 percent stake and making it Mistral’s largest shareholder. The deal also gives ASML a seat on Mistral’s strategic committee. While Mistral lags behind Silicon Valley’s AI giants in terms of capital and speed of innovation, it remains Europe’s major AI unicorn. ASML, meanwhile, is singular: The Dutch firm is the only company that makes the most advanced extreme ultraviolet (EUV) lithography machines used in semiconductor manufacturing. The partnership can be read as a powerful affirmation of European AI sovereignty.
The chatbot’s take: For those who are very much not at risk of being poached for their AI expertise, here’s an explainer about SK Hynix’s memory breakthrough:
–
By Alex Perala
